Nicolas Zucchet

Postdoctoral researcher at Stanford

prof_pic.jpg

I am a postdoctoral researcher at Stanford University in Scott Linderman’s group, supported by a SNSF Postdoc Mobility fellowship. I received my PhD from ETH Zürich in 2025 under the supervision of João Sacramento and Angelika Steger. During my PhD, I was a student researcher at Google DeepMind in London, working with Soham De. I earned my Master’s degree in computer science from ETH Zürich in 2021, following a Bachelor’s degree in Applied Mathematics from École Polytechnique in 2019.

My research interests lie at the intersection of artificial intelligence and neuroscience. I use artificial neural networks to gain insights into how our brains learn and why biological systems are more energy and data efficient. A key focus of my work is developing biophysically plausible learning algorithms for (recurrent) neural networks. This research helps us understand the mechanisms underlying learning in the brain and could reduce the energetic cost of AI systems by bringing neural computation closer to its physical substrate. I am also interested in how data shapes learning, specifically investigating how data curricula could accelerate learning in neural networks.

Outside research, I enjoy road cycling and photography.


News

Oct 15, 2026 Our curse of ambiguity paper was accepted as a spotlight at NeurIPS. See you in Paris!
Sep 15, 2026 I gave a tutorial on scaling laws and how we can understand them using toy models at the Tübingen Machine Learning Summer School (slides)!
Feb 17, 2026 New blog post on a fun new algorithm questioning the fundamentals of learning in recurrent networks.
Feb 12, 2026 My PhD work was recognized with an award from the Swiss computational neuroscience community!
Jan 12, 2026 I am starting a postdoc in Scott Linderman’s group at Stanford!

Selected publications

  1. Language models suffer from a curse of ambiguity
    Nicolas Zucchet, Hyun Dong Lee, and Scott Linderman
    In Advances in Neural Information Processing Systems, 2026
  2. The emergence of sparse attention: impact of data distribution and benefits of repetition
    Nicolas Zucchet, Francesco d’Angelo, Andrew Lampinen, and Stephanie Chan
    Advances in Neural Information Processing Systems, 2025
  3. How do language models learn facts? Dynamics, curricula and hallucinations
    Nicolas Zucchet, Jörg Bornschein, Stephanie Chan, Andrew Lampinen, Razvan Pascanu, and Soham De
    Conference on Language Modeling, 2025
  4. Recurrent neural networks: vanishing and exploding gradients are not the end of the story
    Nicolas Zucchet, and Antonio Orvieto
    Advances in Neural Information Processing Systems, 2024